Guardado en:
Detalles Bibliográficos
Autores principales: Lu, Xinyang, Niu, Xinyuan, Lau, Gregory Kang Ruey, Nhung, Bui Thi Cam, Sim, Rachael Hwee Ling, Himawan, John Russell, Wen, Fanyu, Foo, Chuan-Sheng, Ng, See-Kiong, Low, Bryan Kian Hsiang
Formato: Preprint
Publicado: 2025
Materias:
Acceso en línea:https://arxiv.org/abs/2505.05064
Etiquetas: Agregar Etiqueta
Sin Etiquetas, Sea el primero en etiquetar este registro!
_version_ 1866908803578986496
author Lu, Xinyang
Niu, Xinyuan
Lau, Gregory Kang Ruey
Nhung, Bui Thi Cam
Sim, Rachael Hwee Ling
Himawan, John Russell
Wen, Fanyu
Foo, Chuan-Sheng
Ng, See-Kiong
Low, Bryan Kian Hsiang
author_facet Lu, Xinyang
Niu, Xinyuan
Lau, Gregory Kang Ruey
Nhung, Bui Thi Cam
Sim, Rachael Hwee Ling
Himawan, John Russell
Wen, Fanyu
Foo, Chuan-Sheng
Ng, See-Kiong
Low, Bryan Kian Hsiang
contents Large language model (LLM) unlearning is critical in real-world applications where it is necessary to efficiently remove the influence of private, copyrighted, or harmful data from some users. Existing utility-centric unlearning metrics (based on model utility) may fail to accurately evaluate the extent of unlearning in realistic settings such as when the forget and retain sets have semantically similar content and/or retraining the model from scratch on the retain set is impractical. This paper presents the first data-centric unlearning metric for LLMs called WaterDrum that exploits robust text watermarking to overcome these limitations. We introduce new benchmark datasets (with different levels of data similarity) for LLM unlearning that can be used to rigorously evaluate unlearning algorithms via WaterDrum. Our code is available at https://github.com/lululu008/WaterDrum and our new benchmark datasets are released at https://huggingface.co/datasets/Glow-AI/WaterDrum-Ax.
format Preprint
id arxiv_https___arxiv_org_abs_2505_05064
institution arXiv
publishDate 2025
record_format arxiv
spellingShingle WaterDrum: Watermarking for Data-centric Unlearning Metric
Lu, Xinyang
Niu, Xinyuan
Lau, Gregory Kang Ruey
Nhung, Bui Thi Cam
Sim, Rachael Hwee Ling
Himawan, John Russell
Wen, Fanyu
Foo, Chuan-Sheng
Ng, See-Kiong
Low, Bryan Kian Hsiang
Machine Learning
Large language model (LLM) unlearning is critical in real-world applications where it is necessary to efficiently remove the influence of private, copyrighted, or harmful data from some users. Existing utility-centric unlearning metrics (based on model utility) may fail to accurately evaluate the extent of unlearning in realistic settings such as when the forget and retain sets have semantically similar content and/or retraining the model from scratch on the retain set is impractical. This paper presents the first data-centric unlearning metric for LLMs called WaterDrum that exploits robust text watermarking to overcome these limitations. We introduce new benchmark datasets (with different levels of data similarity) for LLM unlearning that can be used to rigorously evaluate unlearning algorithms via WaterDrum. Our code is available at https://github.com/lululu008/WaterDrum and our new benchmark datasets are released at https://huggingface.co/datasets/Glow-AI/WaterDrum-Ax.
title WaterDrum: Watermarking for Data-centric Unlearning Metric
topic Machine Learning
url https://arxiv.org/abs/2505.05064